Article

Google equips Genie 4 with a Gemini “director”: generating worlds while planning narratives

Beating AI news flash: Google DeepMind has released its new-generation world model, Genie 4. Text input can generate explorable virtual worlds, with keyboard-controlled movement and the ability to change the weather or add characters at any time. It can simulate collisions, aerodynamics, and character movements, and for the first time natively generates ambient sound, music, and real-time dialogue. Characters can not only speak, but also respond based on the scene and user input.

Genie 4 uses Gemini Omni to enhance the realism of visuals and real-world behavior, while Gemini 3.8 Flash serves as a real-time director. The latter continuously observes the screen, breaks tasks into steps, checks whether actions have been completed, and then decides the next step. Google also demonstrated Gemini 4 Argon planning for a humanoid robot to play checkers, with Genie 4 simulating the actions in real time.

A single simulation can last up to approximately 2 minutes, twice the duration of the previous Project Genie experience. However, when a scene runs for a long time, object positions and appearances may still change. In the official demonstration, a motorcycle rider unexpectedly gained a helmet between shots.

Genie 4 is currently available only to a small number of invited testers, and will subsequently be gradually introduced to Google AI Ultra、Gemini API and AI Studio. Google hopes to use it for game prototyping and film-scene previsualization and, more importantly, to enable robots and AI Agent to practice complex tasks in virtual worlds.

Original link https://m.theblockbeats.info/flash/374063